Re: [PATCH net-next] net: dsa: validate that DT nodes of shared ports have the properties they need

[Vladimir Oltean <vladimir.oltean@nxp.com>]

On Sat, Jul 23, 2022 at 08:09:34PM +0200, Andrew Lunn wrote:
> Hi Vladimir
> 
> I think this is a first good step.
> 
> > +static const char * const dsa_switches_skipping_validation[] = {
> 
> One thing to consider is do we want to go one step further and make
> this dsa_switches_dont_enforce_validation
> 
> Meaning always run the validation, and report what is not valid, but
> don't enforce with an -EINVAL for switches on the list?

Can do. The question is what are our prospects of eventually getting rid
of incompletely specified DT blobs. If they're likely to stay around
forever, maybe printing with dev_err() doesn't sound like such a great
idea?

I know what's said in Documentation/devicetree/bindings/net/dsa/marvell.txt
about not putting a DT blob somewhere where you can't update it, but
will that be the case for everyone? Florian, I think some bcm_sf2 device
trees are pretty much permanent, based on some of your past commits?

> Maybe it is too early for that, we first need to submit patches to the
> mainline DT files to fixes those we can?
> 
> Looking at the mv88e6xxx instances, adding fixed-links should not be
> too hard. What might be harder is the phy-mode, in particular, what
> RGMII delay should be specified.

Since DT blobs and kernels have essentially separate lifetimes, I'm
thinking it doesn't matter too much if we first fix the mainline DT
blobs or not; it's not like that would avoid users seeing errors.

Anyway I'm thinking it would be useful in general to verbally resolve
some of the incomplete DT descriptions I've pointed out here. This would
be a good indication whether we can add automatic logic that comes to
the same resolution at least for all known cases.